Search papers, labs, and topics across Lattice.
This paper introduces MixCompress, a novel variable bit-rate (VBR) framework for learned image compression that leverages sparse structural specialization to overcome the limitations of existing methods. By employing a Mixture-of-Experts (MoE) routing mechanism and a Mixture-of-Depths (MoD) extension, the framework dynamically adjusts model capacity to effectively manage gradient conflicts and enhance detail preservation across varying bit rates. Extensive evaluations show that MixCompress not only matches but also surpasses traditional single-rate baselines, setting a new standard for computational efficiency in image coding.
MixCompress redefines the landscape of learned image compression by dynamically scaling model capacity to achieve superior performance across variable bit rates.
Learned image compression (LIC) is bottlenecked by the need to store independent models for each rate-distortion operating point. Existing variable bit-rate (VBR) methods aim to reduce this overhead via dense parameter modulation, but forcing a shared backbone to approximate divergent mappings causes severe feature entanglement. Specifically, low-rate smoothing gradients inherently conflict with the preservation of high-frequency textural details, leading to sub-optimal performance. To resolve this, we propose MixCompress, a unified VBR framework based on sparse structural specialization. While sparsely gated Mixture-of-Experts (MoE) routing successfully mitigates gradient conflict, it operates on a fixed computational budget. To address the increased representational demands of higher bit-rates we introduce a Mixture-of-Depths (MoD) extension to dynamically scale model capacity. Combined with Conditional Auxiliary Transforms (CAT) for dynamic sub-band energy modulation, our hierarchical framework effectively dynamically scales capacity. Extensive evaluations demonstrate that MixCompress not only matches individually optimized single-rate baselines but can even surpass them, establishing a new Pareto frontier for computationally efficient image coding.
